Estimação de Variância Latente no Modelo de Heston via Filtragem Gaussiana

This paper proposes a Gaussian filter to estimate the latent variance in the Heston model discretized by the Euler-Maruyama method. The approach exploits analytical expressions for the moments required by the filtering recursion, avoiding sampling-based calculations such as those used in the Unscented Kalman Filter. In simulations, the proposed method achieves a lower mean squared error than an exponentially weighted moving average estimator. The variance estimates are then converted into absolute price volatility and used to compute bid and ask quotes in the Avellaneda-Stoikov model, using real Amazon stock data from the LOBSTER database.